A career-plan platform for Brazilian developers. Three Claude Opus 4.7 agents read you, crawl the world, and write a personal 90-day plan around the grants, fellowships, and bolsas you didn't know existed, and didn't know you'd be a fit for.
Live demo: radar.pabloaa.com Demo video (3 min): https://youtu.be/ueLPzXevysQ License: AGPL-3.0 Built for: Cerebral Valley Built with Opus 4.7 hackathon, April 2026
Of the ~500 developers approved for this hackathon, only 2 are Brazilian. The same ratio shows up across fellowships, grants, and accelerators. Brazil has over a million developers. The talent isn't missing. It's unmapped.
I opened a waitlist this week. Fifty Brazilian developers signed up in five days. I sat down with five of them. Different stages, same gap: an undergraduate with published research and mentors who couldn't name a single fellowship he was a strong fit for; a senior mid-career engineer looking for what's next; a junior dev who got her first job through a referral and had never heard of any of this. Not "I don't know what exists", but "I don't know what I'd be a fit for".
Brazil doesn't have a brain drain. It has an information drain. Radar finds the opportunity, and finds the person.
I went to Japan twice through METI, the Ministry of Economy, Trade and Industry. I won a grant that put me at Harvard and MIT for an immersion last year. None of it came from a system. I found each opportunity by chance. Every day, on a community of about 60k Brazilian developers I run on Instagram (@pablo_aa), the same DM lands: "where do you find these things?". Radar turns that DM into a product.
Three composed agents on Claude Opus 4.7:
- Anamnesis reads your GitHub, an optional CV, and a personal site URL. It builds a structured profile plus a rich editorial self-portrait: archetype, peer constellation, year-shape, three-year vectors, recommended readings.
- Scout is one shared long-running Managed Agent session that crawls 1,240+ curated sources weekly (FAPESP, Emergent Ventures, Chevening, MEXT, Y Combinator, Fundação Estudar, GSoC, MATS, ...) and persists structured opportunities to Supabase. At every run it discovers and queues 2-5 adjacent sources for the next pass.
- Strategist runs per user, applies eliminatory clarify-answer filters (relocation appetite, time budget, study appetite, ambition vector), ranks opportunities against the profile, writes a why-you paragraph that cites specific profile fields, and produces a 90-day plan tied to each card.
Scout and Strategist run on the Anthropic Managed Agents API (beta), with sessions reused across runs, custom tools (render_card, query_opps, upsert_opportunity, mark_discarded, suggest_source), and per-user memory on Strategist. Anamnesis runs on the Messages API with streaming and 32k-token outputs, because the editorial render does not benefit from session memory and the PDF document-block flow was simpler on plain Messages. See docs/03-managed-agents-deep-dive.md for the full reasoning.
See docs/02-architecture.md for the data model and component picture, and docs/03-managed-agents-deep-dive.md for the design decisions behind the Managed Agents composition.
radar/
├── src/ Next.js 16 App Router, landing + auth + agent UIs
├── public/ static assets
├── supabase/ schema migrations and config
├── scripts/ CLI tooling (Anamnesis, Scout, Strategist setup)
├── data/sample/ fake profiles for tests and public demos
└── docs/ product brief, architecture, managed agents deep-dive
- Production deployment at radar.pabloaa.com
- GitHub OAuth via Supabase, RLS policies per table
- Anamnesis end-to-end with PDF CV ingestion, streaming 32k outputs, defensive JSON parsing
- Scout running on a GitHub Actions cron, single-source protocol with discovery mandate, 1,240+ sources indexed
- Strategist with 4-section output (dated, recurrent, rolling, arenas) plus 90-day plan, eliminatory clarify-answer filters, and BR institutional context (MEI/PJ, FAPESP, USD-BRL conversion)
- Resend email notifications when agent runs complete
- Pre-commit hook blocking secret leaks (sk-ant, ghp_, JWT shapes)
- Rule-based scoring fallback when Strategist errors
- Not a job board. LinkedIn and the established job aggregators already solve that. Radar's universe is grants, fellowships, bolsas, accelerators, OSS funding, bounties, and exchange programs.
- Not auto-apply. Radar surfaces, explains, and strategizes. Applying is always the user's decision.
- Not resume optimization. Dozens of tools do that already.
AGPL-3.0-or-later. Any hosted fork must share its source under the same license.